XOM low oil-beta phases: next-20d performance vs SPY after bottom-quintile Brent correlation
Low oil beta has been a decent place to own Exxon rather than the index. Over the past three years, whenever XOM's rolling 20-day correlation to Brent crude dropped into the bottom fifth of its trailing-year range, the stock beat SPY by an average of 2.25 percentage points over the following 20 sessions — +3.14% versus +0.88%. Nearly 70% of those windows ended with XOM ahead, against roughly 43% on ordinary days, when XOM actually lagged slightly.
The idea being tested is that when crude correlation collapses, the supermajor starts trading on capital-return news rather than oil beta, and that re-rating shows up in relative performance. The sample leans that way, with a caveat worth taking seriously: clustered low-correlation spells mean the 138 events overlap, so the effective sample is smaller than the headline count suggests.
The full breakdown follows.
Over the past ~3 years, when the rolling 20-day correlation between XOM's daily returns and Brent crude returns falls into the bottom quintile of its prior 252-session distribution, does XOM outperform SPY over the next 20 trading sessions? I expect low realized crude correlation to mark an idiosyncratic re-rating phase where the supermajor trades on capital-return signals rather than oil beta, so XOM leads broad equities as the decoupling persists.
How this was measured
Daily closes were rebuilt from XOM and SPY minute bars. Brent returns were taken from brent_daily_df with a one-session lag to respect T+1 macro publication timing. For each session I computed the rolling 20-day Pearson correlation between XOM returns and lagged Brent returns. The condition fires when the current correlation sits at or below the 20th percentile of the prior 252 sessions of the same rolling-correlation series, with the prior window shifted by one session so the current value is not in its own reference distribution. Forward performance is simple close-to-close over the next 20 trading sessions, measured as XOM return minus SPY return.
The key numbers
Reading the numbers
In plain terms: when XOM's 20-day correlation to Brent was in the bottom fifth of its prior 252-session range, XOM beat SPY by +2.25% on average over the next 20 sessions, versus -0.61% on other days. The 69.6% hit rate and p=0.0016 point to a conditional edge, though overlapping windows make the inference nominal.
The charts
This line chart tracks XOM's rolling 20-day correlation to Brent against the 20th-percentile threshold that defines low oil beta. The key is the threshold itself: it starts near -0.16 and slides to -0.357 by the end, so an event is triggered only when the current correlation drops below an already-low and falling bar. The correlation line ranges from -0.68 to +0.63 and averages -0.038, so the 138 event days are the moments when XOM's oil link is unusually weak for that regime; those are the setups the rest of the report tests.
This box plot compares the next-20-day XOM-minus-SPY excess return after bottom-quintile correlation events (138 cases) with all non-event days (321 cases). The event group's mean is +2.25%, while non-event days average -0.61%; the event box therefore sits higher, showing that the average forward edge is concentrated in the low-correlation regime. The min and max ranges still overlap heavily (-26.5% to +22.3% for events, -17.7% to +25.5% for non-events), so this is a shift in the distribution, not a clean win every time.
This histogram shows the spread of the 138 event-day outcomes for XOM minus SPY over the following 20 sessions. The distribution is centered on a positive mean of +2.25%, and the 69.6% positive hit rate says more event windows landed above zero than below. But the tails are wide, from -26.5% to +22.3%, so the chart supports an average conditional edge rather than a guarantee; the large negative tail is where XOM badly lagged despite low oil correlation.
Forward 20-session return summary
| Group | N | Mean XOM fwd20 | Mean SPY fwd20 | Mean XOM-minus-SPY | Share positive |
|---|---|---|---|---|---|
| Bottom-quintile corr events | 138 | 0.0314 | 0.0088 | 0.0225 | 0.6957 |
| Non-event days | 321 | 0.0103 | 0.0164 | -0.0061 | 0.4268 |
| All condition-measurable days | 459 | 0.0166 | 0.0141 | 0.0025 | 0.5076 |
The takeaway
Yes — on this sample, low crude-correlation stretches were followed by XOM beating SPY. Over the 138 sessions where XOM's 20-day correlation to Brent sat in the bottom fifth of its trailing year, XOM averaged +3.14% over the next 20 sessions against SPY's +0.88%, a +2.25% edge. It wasn't a couple of lucky windows carrying it: nearly 70% of those 138 events ended with XOM ahead, versus about 43% on ordinary days. Outside the events, XOM actually lagged slightly (-0.61% excess), so the spread between the two regimes is about +2.86%. The gap is statistically hard to dismiss on paper — a t-stat of 3.22 and a p-value of 0.0016 on the event excess, with an even stronger 0.0006 when events are pitted directly against non-events. Treat that confidence as a notch or two lower than it looks, though: the 20-day windows overlap heavily when low-correlation spells cluster, so the data behaves like fewer independent observations than 138 suggests. Net read: a real, economically meaningful lean in the direction you expected, but a supportive one rather than a proven edge. The practical takeaway is that the decoupling phase has historically been a decent spot to hold XOM over SPY on a 20-day view — not a mechanical signal to size aggressively on, and one that lives entirely within a fairly short sample.
The fine print
- Overlapping 20-day windows mean the 138 events aren't 138 independent bets; the p-values are nominal and flatter the result. Clustered events are effectively one call repeated.
- Only ~459 sessions were even measurable, and the first signal can't fire until 272 sessions of history exist — so this is a short, single-regime sample.
- The bottom-quintile cutoff is XOM's own in-sample historical percentile, not an absolute crude-beta level, so the 'low correlation' bar drifts with the sample.
- Brent is lagged one session for T+1 publication timing, and results are XOM-specific — other supermajors or sharply different oil regimes may behave differently.